loading
loading
Everything you do to a pretrained base model to make it helpful, safe, and good at following instructions.
Post-training is the umbrella for the stages after pretraining: supervised fine-tuning on instruction/response pairs, then preference alignment like RLHF or DPO, plus safety and reasoning training. It's far cheaper than pretraining but it's what turns a raw text-completer into something like Claude that actually answers your question instead of autocompleting it. The base model holds the knowledge; post-training shapes the behavior — tone, refusals, formatting, tool use.
Plainly
Think of Post-training as the brain part that guesses or decides. Everything you do to a pretrained base model to make it helpful, safe, and good at following instructions.
In practice
Use it when model choice, prompts, latency, cost, or quality affect the product result. In practice, define the owner, input, output, and failure mode before you rely on it.